static void r300_emit_draw_elements(struct r300_context *r300,
struct pipe_buffer* indexBuffer,
unsigned indexSize,
unsigned minIndex,
unsigned maxIndex,
unsigned mode,
unsigned start,
unsigned count)
{
uint32_t count_dwords;
uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
CS_LOCALS(r300);
/* XXX most of these are stupid */
assert(indexSize == 4 || indexSize == 2);
assert((start * indexSize) % 4 == 0);
assert(offset_dwords == 0);
BEGIN_CS(10);
OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, maxIndex);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
if (indexSize == 4) {
count_dwords = count + start;
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
r300_translate_primitive(mode));
} else {
count_dwords = (count + start + 1) / 2;
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
r300_translate_primitive(mode));
}
/* INDX_BUFFER is a truly special packet3.
* Unlike most other packet3, where the offset is after the count,
* the order is reversed, so the relocation ends up carrying the
* size of the indexbuf instead of the offset.
*
* XXX Fix offset
*/
OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
(0 << R300_INDX_BUFFER_SKIP_SHIFT));
OUT_CS(offset_dwords);
OUT_CS_RELOC(indexBuffer, count_dwords,
RADEON_GEM_DOMAIN_GTT, 0, 0);
END_CS;
}
void r300_emit_vertex_arrays_swtcl(struct r300_context *r300, boolean indexed)
{
CS_LOCALS(r300);
DBG(r300, DBG_SWTCL, "r300: Preparing vertex buffer %p for render, "
"vertex size %d\n", r300->vbo,
r300->vertex_info.size);
/* Set the pointer to our vertex buffer. The emitted values are this:
* PACKET3 [3D_LOAD_VBPNTR]
* COUNT [1]
* FORMAT [size | stride << 8]
* OFFSET [offset into BO]
* VBPNTR [relocated BO]
*/
BEGIN_CS(7);
OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, 3);
OUT_CS(1 | (!indexed ? R300_VC_FORCE_PREFETCH : 0));
OUT_CS(r300->vertex_info.size |
(r300->vertex_info.size << 8));
OUT_CS(r300->draw_vbo_offset);
OUT_CS(0);
assert(r300->vbo_cs);
OUT_CS(0xc0001000); /* PKT3_NOP */
OUT_CS(r300->rws->cs_get_reloc(r300->cs, r300->vbo_cs) * 4);
END_CS;
}
static void r300_emit_draw_arrays(struct r300_context *r300,
unsigned mode,
unsigned count)
{
boolean alt_num_verts = count > 65535;
CS_LOCALS(r300);
if (count >= (1 << 24)) {
fprintf(stderr, "r300: Got a huge number of vertices: %i, "
"refusing to render.\n", count);
return;
}
r300_emit_draw_init(r300, mode, count-1);
BEGIN_CS(2 + (alt_num_verts ? 2 : 0));
if (alt_num_verts) {
OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
}
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
r300_translate_primitive(mode) |
(alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
END_CS;
}
static void r300_render_draw_arrays(struct vbuf_render* render,
unsigned start,
unsigned count)
{
struct r300_render* r300render = r300_render(render);
struct r300_context* r300 = r300render->r300;
uint8_t* ptr;
unsigned i;
unsigned dwords = 6;
CS_LOCALS(r300);
(void) i; (void) ptr;
assert(start == 0);
assert(count < (1 << 16));
DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count);
if (!r300_prepare_for_rendering(r300,
PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL,
NULL, dwords, 0, 0, -1)) {
return;
}
BEGIN_CS(dwords);
OUT_CS_REG(R300_GA_COLOR_CONTROL,
r300_provoking_vertex_fixes(r300, r300render->prim));
OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
r300render->hwprim);
END_CS;
}
static void r300_render_draw_elements(struct vbuf_render* render,
const ushort* indices,
uint count)
{
struct r300_render* r300render = r300_render(render);
struct r300_context* r300 = r300render->r300;
unsigned max_index = (r300->vbo->size - r300->draw_vbo_offset) /
(r300render->r300->vertex_info.size * 4) - 1;
struct pipe_resource *index_buffer = NULL;
unsigned index_buffer_offset;
CS_LOCALS(r300);
DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count);
u_upload_data(r300->uploader, 0, count * 2, indices,
&index_buffer_offset, &index_buffer);
if (!index_buffer) {
return;
}
if (!r300_prepare_for_rendering(r300,
PREP_EMIT_STATES |
PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
index_buffer, 12, 0, 0, -1)) {
pipe_resource_reference(&index_buffer, NULL);
return;
}
BEGIN_CS(12);
OUT_CS_REG(R300_GA_COLOR_CONTROL,
r300_provoking_vertex_fixes(r300, r300render->prim));
OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
r300render->hwprim);
OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2));
OUT_CS(index_buffer_offset);
OUT_CS((count + 1) / 2);
OUT_CS_RELOC(r300_resource(index_buffer));
END_CS;
pipe_resource_reference(&index_buffer, NULL);
}
static void r300_render_draw_arrays(struct vbuf_render* render,
unsigned start,
unsigned count)
{
struct r300_render* r300render = r300_render(render);
struct r300_context* r300 = r300render->r300;
uint8_t* ptr;
unsigned i;
unsigned dwords = 6;
CS_LOCALS(r300);
(void) i; (void) ptr;
DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count);
if (r300->draw_first_emitted) {
if (!r300_prepare_for_rendering(r300,
PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL,
NULL, 6, 0, 0))
return;
} else {
if (!r300_emit_states(r300,
PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL,
NULL, 0, 0))
return;
}
/* Uncomment to dump all VBOs rendered through this interface.
* Slow and noisy!
ptr = pipe_buffer_map(&r300render->r300->context,
r300render->vbo, PIPE_TRANSFER_READ,
&r300render->vbo_transfer);
for (i = 0; i < count; i++) {
printf("r300: Vertex %d\n", i);
draw_dump_emitted_vertex(&r300->vertex_info, ptr);
ptr += r300->vertex_info.size * 4;
printf("\n");
}
pipe_buffer_unmap(&r300render->r300->context, r300render->vbo,
r300render->vbo_transfer);
*/
BEGIN_CS(dwords);
OUT_CS_REG(R300_GA_COLOR_CONTROL,
r300_provoking_vertex_fixes(r300, r300render->prim));
OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
r300render->hwprim);
END_CS;
r300->draw_first_emitted = TRUE;
}
static void r300_emit_draw_arrays(struct r300_context *r300,
unsigned mode,
unsigned count)
{
CS_LOCALS(r300);
BEGIN_CS(4);
OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
r300_translate_primitive(mode));
END_CS;
}
static void r300_render_draw_arrays(struct vbuf_render* render,
unsigned start,
unsigned count)
{
struct r300_render* r300render = r300_render(render);
struct r300_context* r300 = r300render->r300;
CS_LOCALS(r300);
r300_emit_dirty_state(r300);
DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
BEGIN_CS(2);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
r300render->hwprim);
END_CS;
}
void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state)
{
struct pipe_framebuffer_state *fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
struct r300_resource *tex;
CS_LOCALS(r300);
tex = r300_resource(fb->zsbuf->texture);
BEGIN_CS(size);
OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_ZMASK, 2);
OUT_CS(0);
OUT_CS(tex->tex.zmask_dwords[fb->zsbuf->u.tex.level]);
OUT_CS(0);
END_CS;
/* Mark the current zbuffer's zmask as in use. */
r300->zmask_in_use = TRUE;
r300_mark_atom_dirty(r300, &r300->hyperz_state);
}
void r300_emit_cmask_clear(struct r300_context *r300, unsigned size, void *state)
{
struct pipe_framebuffer_state *fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
struct r300_resource *tex;
CS_LOCALS(r300);
tex = r300_resource(fb->cbufs[0]->texture);
BEGIN_CS(size);
OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_CMASK, 2);
OUT_CS(0);
OUT_CS(tex->tex.cmask_dwords);
OUT_CS(0);
END_CS;
/* Mark the current zbuffer's zmask as in use. */
r300->cmask_in_use = TRUE;
r300_mark_fb_state_dirty(r300, R300_CHANGED_CMASK_ENABLE);
}
void r300_emit_hiz_clear(struct r300_context *r300, unsigned size, void *state)
{
struct pipe_framebuffer_state *fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
struct r300_resource* tex;
CS_LOCALS(r300);
tex = r300_resource(fb->zsbuf->texture);
BEGIN_CS(size);
OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT,
R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_HIZ, 2);
OUT_CS(0);
OUT_CS(tex->tex.hiz_dwords[fb->zsbuf->u.tex.level]);
OUT_CS(r300->hiz_clear_value);
END_CS;
/* Mark the current zbuffer's hiz ram as in use. */
r300->hiz_in_use = TRUE;
r300->hiz_func = HIZ_FUNC_NONE;
r300_mark_atom_dirty(r300, &r300->hyperz_state);
}
static void r300_render_draw(struct vbuf_render* render,
const ushort* indices,
uint count)
{
struct r300_render* r300render = r300_render(render);
struct r300_context* r300 = r300render->r300;
int i;
CS_LOCALS(r300);
r300_emit_dirty_state(r300);
BEGIN_CS(2 + (count+1)/2);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
r300render->hwprim);
for (i = 0; i < count-1; i += 2) {
OUT_CS(indices[i+1] << 16 | indices[i]);
}
if (count % 2) {
OUT_CS(indices[count-1]);
}
END_CS;
}
static void r300_draw_elements_immediate(struct r300_context *r300,
const struct pipe_draw_info *info)
{
const uint8_t *ptr1;
const uint16_t *ptr2;
const uint32_t *ptr4;
unsigned index_size = r300->index_buffer.index_size;
unsigned i, count_dwords = index_size == 4 ? info->count :
(info->count + 1) / 2;
CS_LOCALS(r300);
/* 19 dwords for r300_draw_elements_immediate. Give up if the function fails. */
if (!r300_prepare_for_rendering(r300,
PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS |
PREP_INDEXED, NULL, 2+count_dwords, 0, info->index_bias, -1))
return;
r300_emit_draw_init(r300, info->mode, info->max_index);
BEGIN_CS(2 + count_dwords);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, count_dwords);
switch (index_size) {
case 1:
ptr1 = (uint8_t*)r300->index_buffer.user_buffer;
ptr1 += info->start;
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
r300_translate_primitive(info->mode));
if (info->index_bias && !r300->screen->caps.is_r500) {
for (i = 0; i < info->count-1; i += 2)
OUT_CS(((ptr1[i+1] + info->index_bias) << 16) |
(ptr1[i] + info->index_bias));
if (info->count & 1)
OUT_CS(ptr1[i] + info->index_bias);
} else {
for (i = 0; i < info->count-1; i += 2)
OUT_CS(((ptr1[i+1]) << 16) |
(ptr1[i] ));
if (info->count & 1)
OUT_CS(ptr1[i]);
}
break;
case 2:
ptr2 = (uint16_t*)r300->index_buffer.user_buffer;
ptr2 += info->start;
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
r300_translate_primitive(info->mode));
if (info->index_bias && !r300->screen->caps.is_r500) {
for (i = 0; i < info->count-1; i += 2)
OUT_CS(((ptr2[i+1] + info->index_bias) << 16) |
(ptr2[i] + info->index_bias));
if (info->count & 1)
OUT_CS(ptr2[i] + info->index_bias);
} else {
OUT_CS_TABLE(ptr2, count_dwords);
}
break;
case 4:
ptr4 = (uint32_t*)r300->index_buffer.user_buffer;
ptr4 += info->start;
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
r300_translate_primitive(info->mode));
if (info->index_bias && !r300->screen->caps.is_r500) {
for (i = 0; i < info->count; i++)
OUT_CS(ptr4[i] + info->index_bias);
} else {
OUT_CS_TABLE(ptr4, count_dwords);
}
break;
}
END_CS;
}
static void r300_emit_draw_elements(struct r300_context *r300,
struct pipe_resource* indexBuffer,
unsigned indexSize,
unsigned max_index,
unsigned mode,
unsigned start,
unsigned count,
uint16_t *imm_indices3)
{
uint32_t count_dwords, offset_dwords;
boolean alt_num_verts = count > 65535;
CS_LOCALS(r300);
if (count >= (1 << 24)) {
fprintf(stderr, "r300: Got a huge number of vertices: %i, "
"refusing to render (max_index: %i).\n", count, max_index);
return;
}
DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, max %u\n",
count, max_index);
r300_emit_draw_init(r300, mode, max_index);
/* If start is odd, render the first triangle with indices embedded
* in the command stream. This will increase start by 3 and make it
* even. We can then proceed without a fallback. */
if (indexSize == 2 && (start & 1) &&
mode == PIPE_PRIM_TRIANGLES) {
BEGIN_CS(4);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 2);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (3 << 16) |
R300_VAP_VF_CNTL__PRIM_TRIANGLES);
OUT_CS(imm_indices3[1] << 16 | imm_indices3[0]);
OUT_CS(imm_indices3[2]);
END_CS;
start += 3;
count -= 3;
if (!count)
return;
}
offset_dwords = indexSize * start / sizeof(uint32_t);
BEGIN_CS(8 + (alt_num_verts ? 2 : 0));
if (alt_num_verts) {
OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
}
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
if (indexSize == 4) {
count_dwords = count;
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
r300_translate_primitive(mode) |
(alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
} else {
count_dwords = (count + 1) / 2;
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
r300_translate_primitive(mode) |
(alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
}
OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
(0 << R300_INDX_BUFFER_SKIP_SHIFT));
OUT_CS(offset_dwords << 2);
OUT_CS(count_dwords);
OUT_CS_RELOC(r300_resource(indexBuffer));
END_CS;
}
static void r300_draw_arrays_immediate(struct r300_context *r300,
const struct pipe_draw_info *info)
{
struct pipe_vertex_element* velem;
struct pipe_vertex_buffer* vbuf;
unsigned vertex_element_count = r300->velems->count;
unsigned i, v, vbi;
/* Size of the vertex, in dwords. */
unsigned vertex_size = r300->velems->vertex_size_dwords;
/* The number of dwords for this draw operation. */
unsigned dwords = 4 + info->count * vertex_size;
/* Size of the vertex element, in dwords. */
unsigned size[PIPE_MAX_ATTRIBS];
/* Stride to the same attrib in the next vertex in the vertex buffer,
* in dwords. */
unsigned stride[PIPE_MAX_ATTRIBS];
/* Mapped vertex buffers. */
uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
uint32_t* mapelem[PIPE_MAX_ATTRIBS];
CS_LOCALS(r300);
if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
return;
/* Calculate the vertex size, offsets, strides etc. and map the buffers. */
for (i = 0; i < vertex_element_count; i++) {
velem = &r300->velems->velem[i];
size[i] = r300->velems->format_size[i] / 4;
vbi = velem->vertex_buffer_index;
vbuf = &r300->vertex_buffer[vbi];
stride[i] = vbuf->stride / 4;
/* Map the buffer. */
if (!map[vbi]) {
map[vbi] = (uint32_t*)r300->rws->buffer_map(
r300_resource(vbuf->buffer)->cs_buf,
r300->cs, PIPE_TRANSFER_READ | PIPE_TRANSFER_UNSYNCHRONIZED);
map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * info->start;
}
mapelem[i] = map[vbi] + (velem->src_offset / 4);
}
r300_emit_draw_init(r300, info->mode, info->count-1);
BEGIN_CS(dwords);
OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, info->count * vertex_size);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (info->count << 16) |
r300_translate_primitive(info->mode));
/* Emit vertices. */
for (v = 0; v < info->count; v++) {
for (i = 0; i < vertex_element_count; i++) {
OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]);
}
}
END_CS;
}
/* This functions is used to draw a rectangle for the blitter module.
*
* If we rendered a quad, the pixels on the main diagonal
* would be computed and stored twice, which makes the clear/copy codepaths
* somewhat inefficient. Instead we use a rectangular point sprite. */
void r300_blitter_draw_rectangle(struct blitter_context *blitter,
int x1, int y1, int x2, int y2,
float depth,
enum blitter_attrib_type type,
const union pipe_color_union *attrib)
{
struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter));
unsigned last_sprite_coord_enable = r300->sprite_coord_enable;
unsigned width = x2 - x1;
unsigned height = y2 - y1;
unsigned vertex_size =
type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4;
unsigned dwords = 13 + vertex_size +
(type == UTIL_BLITTER_ATTRIB_TEXCOORD ? 7 : 0);
static const union pipe_color_union zeros;
CS_LOCALS(r300);
/* XXX workaround for a lockup in MSAA resolve on SWTCL chipsets, this
* function most probably doesn't handle type=NONE correctly */
if (!r300->screen->caps.has_tcl && type == UTIL_BLITTER_ATTRIB_NONE) {
util_blitter_draw_rectangle(blitter, x1, y1, x2, y2, depth, type, attrib);
return;
}
if (r300->skip_rendering)
return;
if (type == UTIL_BLITTER_ATTRIB_TEXCOORD)
r300->sprite_coord_enable = 1;
r300_update_derived_state(r300);
/* Mark some states we don't care about as non-dirty. */
r300->viewport_state.dirty = FALSE;
if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
goto done;
DBG(r300, DBG_DRAW, "r300: draw_rectangle\n");
BEGIN_CS(dwords);
/* Set up GA. */
OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16));
if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) {
/* Set up the GA to generate texcoords. */
OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE |
(R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT));
OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4);
OUT_CS_32F(attrib->f[0]);
OUT_CS_32F(attrib->f[3]);
OUT_CS_32F(attrib->f[2]);
OUT_CS_32F(attrib->f[1]);
}
/* Set up VAP controls. */
OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT);
OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
OUT_CS(1);
OUT_CS(0);
/* Draw. */
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) |
R300_VAP_VF_CNTL__PRIM_POINTS);
OUT_CS_32F(x1 + width * 0.5f);
OUT_CS_32F(y1 + height * 0.5f);
OUT_CS_32F(depth);
OUT_CS_32F(1);
if (vertex_size == 8) {
if (!attrib)
attrib = &zeros;
OUT_CS_TABLE(attrib->f, 4);
}
END_CS;
done:
/* Restore the state. */
r300_mark_atom_dirty(r300, &r300->rs_state);
r300_mark_atom_dirty(r300, &r300->viewport_state);
r300->sprite_coord_enable = last_sprite_coord_enable;
}
void r300_emit_vertex_arrays(struct r300_context* r300, int offset,
boolean indexed, int instance_id)
{
struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
struct pipe_vertex_element *velem = r300->velems->velem;
struct r300_resource *buf;
int i;
unsigned vertex_array_count = r300->velems->count;
unsigned packet_size = (vertex_array_count * 3 + 1) / 2;
struct pipe_vertex_buffer *vb1, *vb2;
unsigned *hw_format_size = r300->velems->format_size;
unsigned size1, size2, offset1, offset2, stride1, stride2;
CS_LOCALS(r300);
BEGIN_CS(2 + packet_size + vertex_array_count * 2);
OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size);
OUT_CS(vertex_array_count | (!indexed ? R300_VC_FORCE_PREFETCH : 0));
if (instance_id == -1) {
/* Non-instanced arrays. This ignores instance_divisor and instance_id. */
for (i = 0; i < vertex_array_count - 1; i += 2) {
vb1 = &vbuf[velem[i].vertex_buffer_index];
vb2 = &vbuf[velem[i+1].vertex_buffer_index];
size1 = hw_format_size[i];
size2 = hw_format_size[i+1];
OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) |
R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(vb2->stride));
OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
OUT_CS(vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride);
}
if (vertex_array_count & 1) {
vb1 = &vbuf[velem[i].vertex_buffer_index];
size1 = hw_format_size[i];
OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride));
OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
}
for (i = 0; i < vertex_array_count; i++) {
buf = r300_resource(vbuf[velem[i].vertex_buffer_index].buffer);
OUT_CS_RELOC(buf);
}
} else {
/* Instanced arrays. */
for (i = 0; i < vertex_array_count - 1; i += 2) {
vb1 = &vbuf[velem[i].vertex_buffer_index];
vb2 = &vbuf[velem[i+1].vertex_buffer_index];
size1 = hw_format_size[i];
size2 = hw_format_size[i+1];
if (velem[i].instance_divisor) {
stride1 = 0;
offset1 = vb1->buffer_offset + velem[i].src_offset +
(instance_id / velem[i].instance_divisor) * vb1->stride;
} else {
stride1 = vb1->stride;
offset1 = vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride;
}
if (velem[i+1].instance_divisor) {
stride2 = 0;
offset2 = vb2->buffer_offset + velem[i+1].src_offset +
(instance_id / velem[i+1].instance_divisor) * vb2->stride;
} else {
stride2 = vb2->stride;
offset2 = vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride;
}
OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(stride1) |
R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(stride2));
OUT_CS(offset1);
OUT_CS(offset2);
}
if (vertex_array_count & 1) {
vb1 = &vbuf[velem[i].vertex_buffer_index];
size1 = hw_format_size[i];
if (velem[i].instance_divisor) {
stride1 = 0;
offset1 = vb1->buffer_offset + velem[i].src_offset +
(instance_id / velem[i].instance_divisor) * vb1->stride;
} else {
stride1 = vb1->stride;
offset1 = vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride;
}
OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(stride1));
OUT_CS(offset1);
}
for (i = 0; i < vertex_array_count; i++) {
buf = r300_resource(vbuf[velem[i].vertex_buffer_index].buffer);
OUT_CS_RELOC(buf);
}
}
END_CS;
}
/* This functions is used to draw a rectangle for the blitter module.
*
* If we rendered a quad, the pixels on the main diagonal
* would be computed and stored twice, which makes the clear/copy codepaths
* somewhat inefficient. Instead we use a rectangular point sprite. */
static void r300_blitter_draw_rectangle(struct blitter_context *blitter,
unsigned x1, unsigned y1,
unsigned x2, unsigned y2,
float depth,
enum blitter_attrib_type type,
const float attrib[4])
{
struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter));
unsigned last_sprite_coord_enable = r300->sprite_coord_enable;
unsigned width = x2 - x1;
unsigned height = y2 - y1;
unsigned vertex_size =
type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4;
unsigned dwords = 13 + vertex_size +
(type == UTIL_BLITTER_ATTRIB_TEXCOORD ? 7 : 0);
const float zeros[4] = {0, 0, 0, 0};
CS_LOCALS(r300);
if (type == UTIL_BLITTER_ATTRIB_TEXCOORD)
r300->sprite_coord_enable = 1;
r300_update_derived_state(r300);
/* Mark some states we don't care about as non-dirty. */
r300->clip_state.dirty = FALSE;
r300->viewport_state.dirty = FALSE;
if (!r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0))
goto done;
DBG(r300, DBG_DRAW, "r300: draw_rectangle\n");
BEGIN_CS(dwords);
/* Set up GA. */
OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16));
if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) {
/* Set up the GA to generate texcoords. */
OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE |
(R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT));
OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4);
OUT_CS_32F(attrib[0]);
OUT_CS_32F(attrib[3]);
OUT_CS_32F(attrib[2]);
OUT_CS_32F(attrib[1]);
}
/* Set up VAP controls. */
OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT);
OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
OUT_CS(1);
OUT_CS(0);
/* Draw. */
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) |
R300_VAP_VF_CNTL__PRIM_POINTS);
OUT_CS_32F(x1 + width * 0.5f);
OUT_CS_32F(y1 + height * 0.5f);
OUT_CS_32F(depth);
OUT_CS_32F(1);
if (vertex_size == 8) {
if (!attrib)
attrib = zeros;
OUT_CS_TABLE(attrib, 4);
}
END_CS;
done:
/* Restore the state. */
r300->clip_state.dirty = TRUE;
r300->rs_state.dirty = TRUE;
r300->viewport_state.dirty = TRUE;
r300->sprite_coord_enable = last_sprite_coord_enable;
}
static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
unsigned mode,
unsigned start,
unsigned count)
{
struct pipe_vertex_element* velem;
struct pipe_vertex_buffer* vbuf;
unsigned vertex_element_count = r300->velems->count;
unsigned i, v, vbi;
/* Size of the vertex, in dwords. */
unsigned vertex_size = r300->velems->vertex_size_dwords;
/* The number of dwords for this draw operation. */
unsigned dwords = 9 + count * vertex_size;
/* Size of the vertex element, in dwords. */
unsigned size[PIPE_MAX_ATTRIBS];
/* Stride to the same attrib in the next vertex in the vertex buffer,
* in dwords. */
unsigned stride[PIPE_MAX_ATTRIBS];
/* Mapped vertex buffers. */
uint32_t* map[PIPE_MAX_ATTRIBS];
uint32_t* mapelem[PIPE_MAX_ATTRIBS];
struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {0};
CS_LOCALS(r300);
if (!r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0))
return;
/* Calculate the vertex size, offsets, strides etc. and map the buffers. */
for (i = 0; i < vertex_element_count; i++) {
velem = &r300->velems->velem[i];
size[i] = r300->velems->hw_format_size[i] / 4;
vbi = velem->vertex_buffer_index;
vbuf = &r300->vertex_buffer[vbi];
stride[i] = vbuf->stride / 4;
/* Map the buffer. */
if (!transfer[vbi]) {
map[vbi] = (uint32_t*)pipe_buffer_map(&r300->context,
vbuf->buffer,
PIPE_TRANSFER_READ,
&transfer[vbi]);
map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * start;
}
mapelem[i] = map[vbi] + (velem->src_offset / 4);
}
BEGIN_CS(dwords);
OUT_CS_REG(R300_GA_COLOR_CONTROL,
r300_provoking_vertex_fixes(r300, mode));
OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
OUT_CS(count - 1);
OUT_CS(0);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) |
r300_translate_primitive(mode));
/* Emit vertices. */
for (v = 0; v < count; v++) {
for (i = 0; i < vertex_element_count; i++) {
OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]);
}
}
END_CS;
/* Unmap buffers. */
for (i = 0; i < vertex_element_count; i++) {
vbi = r300->velems->velem[i].vertex_buffer_index;
if (transfer[vbi]) {
vbuf = &r300->vertex_buffer[vbi];
pipe_buffer_unmap(&r300->context, vbuf->buffer, transfer[vbi]);
transfer[vbi] = NULL;
}
}
}
static void r300_render_draw_elements(struct vbuf_render* render,
const ushort* indices,
uint count)
{
struct r300_render* r300render = r300_render(render);
struct r300_context* r300 = r300render->r300;
int i;
unsigned end_cs_dwords;
unsigned max_index = (r300->draw_vbo_size - r300->draw_vbo_offset) /
(r300render->r300->vertex_info.size * 4) - 1;
unsigned short_count;
unsigned free_dwords;
CS_LOCALS(r300);
DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count);
if (r300->draw_first_emitted) {
if (!r300_prepare_for_rendering(r300,
PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
NULL, 256, 0, 0, -1))
return;
} else {
if (!r300_emit_states(r300,
PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
NULL, 0, 0, -1))
return;
}
/* Below we manage the CS space manually because there may be more
* indices than it can fit in CS. */
end_cs_dwords = r300_get_num_cs_end_dwords(r300);
while (count) {
free_dwords = RADEON_MAX_CMDBUF_DWORDS - r300->cs->cdw;
short_count = MIN2(count, (free_dwords - end_cs_dwords - 6) * 2);
BEGIN_CS(6 + (short_count+1)/2);
OUT_CS_REG(R300_GA_COLOR_CONTROL,
r300_provoking_vertex_fixes(r300, r300render->prim));
OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (short_count+1)/2);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (short_count << 16) |
r300render->hwprim);
for (i = 0; i < short_count-1; i += 2) {
OUT_CS(indices[i+1] << 16 | indices[i]);
}
if (short_count % 2) {
OUT_CS(indices[short_count-1]);
}
END_CS;
/* OK now subtract the emitted indices and see if we need to emit
* another draw packet. */
indices += short_count;
count -= short_count;
if (count) {
if (!r300_prepare_for_rendering(r300,
PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
NULL, 256, 0, 0, -1))
return;
end_cs_dwords = r300_get_num_cs_end_dwords(r300);
}
}
r300->draw_first_emitted = TRUE;
}
static void r300_emit_draw_elements(struct r300_context *r300,
struct pipe_resource* indexBuffer,
unsigned indexSize,
unsigned minIndex,
unsigned maxIndex,
unsigned mode,
unsigned start,
unsigned count)
{
uint32_t count_dwords;
uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
boolean alt_num_verts = count > 65535;
CS_LOCALS(r300);
if (count >= (1 << 24)) {
fprintf(stderr, "r300: Got a huge number of vertices: %i, "
"refusing to render.\n", count);
return;
}
maxIndex = MIN2(maxIndex, r300->vertex_buffer_max_index);
DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n",
count, minIndex, maxIndex);
BEGIN_CS(13 + (alt_num_verts ? 2 : 0));
if (alt_num_verts) {
OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
}
OUT_CS_REG(R300_GA_COLOR_CONTROL,
r300_provoking_vertex_fixes(r300, mode));
OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
OUT_CS(maxIndex);
OUT_CS(minIndex);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
if (indexSize == 4) {
count_dwords = count;
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
r300_translate_primitive(mode) |
(alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
} else {
count_dwords = (count + 1) / 2;
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
r300_translate_primitive(mode) |
(alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
}
/* INDX_BUFFER is a truly special packet3.
* Unlike most other packet3, where the offset is after the count,
* the order is reversed, so the relocation ends up carrying the
* size of the indexbuf instead of the offset.
*/
OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
(0 << R300_INDX_BUFFER_SKIP_SHIFT));
OUT_CS(offset_dwords << 2);
OUT_CS_BUF_RELOC(indexBuffer, count_dwords,
r300_buffer(indexBuffer)->domain, 0);
END_CS;
}
